Journal of Building Engineering 29 (2020) 101144 Available online 24 December 2019 2352-7102/© 2019 Elsevier Ltd. All rights reserved. Hybrid short-term forecasting of the electric demand of supply fans using machine learning Jason Runge , Radu Zmeureanu * , Mathieu Le Cam Centre for Net-Zero Energy Buildings Studies, Department of Building, Civil and Environmental Engineering, Concordia University, Montreal, Quebec, Canada A R T I C L E INFO Keywords: Multi-step-ahead forecasting Artifcial neural network Ensemble Supply air fow rate Hybrid grey-box Electric demand Sliding window ABSTRACT This paper presents the development and application of multi-step-ahead short-term forecasting models targeting supply fans installed in an institutional building. The models applied in this work consist of an artifcial neural network (ANN) applied in order to forecast the future supply air fow rate of the fans (black-box approach), and a physical model coupled with the ANN applied in order to forecast the future electric demand of the supply fans (hybrid grey-box approach). The forecasting models use measurement data obtained at 15-min intervals in order to forecast the target variables over the next 6 h. The architecture of the ANN was found through an automated search in the training data set. The paper compares the results of selected ANN models with those from other machine learning techniques (support vector regression and ensemble methods) along with a simple forecasting approach. The results of this study show a better forecasting performance when compared with the results from other publications: the CV(RMSE) is 1.8–3.4% for the air fow rate, and 4.8–7.3% for the electric demand for all new models. The results demonstrate that automating the hyperparameter search of the ANN architecture can help alleviate the diffculty of manual parameter setting and achieve a high performing model. 1. Introduction Our cities and populations are growing, concurrently, so is our en- ergy demand. In Canada, approximately 19% of the total secondary energy usage is consumed by electricity [1]. From that, the residential and commercial/institutional sectors consume approximately 53% [1, 2]. Within the United States, the residential and commercial/institu- tional sectors consume approximately 40% of the total primary energy consumption [3] and 71% of the electricity usage [4]. Therefore, increasing the energy effciency in buildings is of great importance to global sustainability. Forecasting the energy consumption in buildings has received a lot of development in recent years as it underpins many techniques for improving building energy performance through: fault detection and diagnosis, demand side management, energy optimization, and demand response. To date, the main focus for forecasting has been on the short- term load profles of buildings as this has close ties to the day-to-day operations. Specifcally, the majority of building energy forecasting papers have focused on forecasting an overall energy load (heating, cooling, and electricity) within a building using hourly data for their models. Despite the focus of forecasting models being applied to the overall energy loads within a building, the heating, ventilation, and air condi- tioning (HVAC) equipment remains an area of interest as it contributes to a large portion of the electric demand within a building (e.g., 19–67% for a commercial building [5]). The HVAC system is a complex, nonlinear system involving numerous variables, many of them correlated. This makes the devel- opment of a forecasting model targeting the electric demand of the HVAC equipment a challenging endeavor, especially as this can be highly occupant driven. This paper is a contribution to the development of short-term forecasting models for the electric demand of HVAC equipment. 2. Literature review 2.1. Overview of forecasting models Merriam-Webster [6] defnes the action to forecast as “to calculate or predict (some future event or condition) usually as a result of study and analysis of available and pertinent data”. Within this paper, to forecast is defned as the estimation of future values of a variable, given its current * Corresponding author. E-mail address: radu.zmeureanu@concordia.ca (R. Zmeureanu). Contents lists available at ScienceDirect Journal of Building Engineering journal homepage: http://www.elsevier.com/locate/jobe https://doi.org/10.1016/j.jobe.2019.101144 Received 19 July 2019; Received in revised form 14 December 2019; Accepted 22 December 2019